LLM hallucination rates will not be brought down to human-expert levels by the end of 2024.
Accurate ●●●●●
“If hallucinations were brought down to human expert levels by the end of 2024 I would be truly astonished.”
- Said on
- 31 March 2024 · Substack post responding to Reid Hoffman's prediction about hallucinations
- Source
- When Will the GenAI Bubble Burst? · Marcus on AI (Substack)
- Should come true
- 2024
- How sure they sounded
- "I would be truly astonished"
- Categories
- AITechnology
- Importance
- 3 of 5 · Reliability of LLMs matters to the whole AI industry.
Outcome
Accurate
Hallucinations were not at human-expert levels by the end of 2024. OpenAI's April 2025 system card showed its newest reasoning models, o3 and o4-mini, hallucinating on 33% and 48% of PersonQA questions, more than earlier models.
- https://www.techzine.eu/news/applications/130720/new-openai-models-hallucinate-more-often-than-their-predecessors/
- https://tech.yahoo.com/articles/openais-o3-o4-mini-hallucinate-173655834.html
History
- 2026-10-17 00:11:44 · created by claude
- 2026-10-17 00:11:44 · categorized by claude
- 2026-10-17 00:12:06 · categorized by claude-new-person-agent
- 2026-10-17 00:12:49 · scored by claude